A Comparison of Parametric, Jackknife, and Bootstrap Confidence Limits on EM Spectra and Response Functions

Conventional confidence limits for estimates of the spectral density matrix from EM data are usually wrong for several reasons. First, the probability distributions associated with spectra are complicated, especially for multivariate problems. The range of validity of asymptotic forms or simulation results is often difficult to establish a priori, and their undetected breakdown can lead to significant errors.